Bayesian analysis of prevalence with covariates using simulation-based techniques: applications to HIV screening

Area of Science:

• Biostatistics
• Epidemiological Methods
• Medical Diagnostic Accuracy

Background:

• Prevalence estimation is biased when diagnostic test precision (sensitivity and specificity) is ignored.
• Treating sensitivity and specificity as constants underestimates prevalence estimate variability.
• Existing Bayesian methods for prevalence estimation with test variability are computationally complex.

Purpose of the Study:

• To extend existing missing-data approaches for prevalence estimation.
• To incorporate covariate effects into prevalence estimation models.
• To generalize Bayesian analysis for binary response data with response errors.

Main Methods:

• Combined a missing-data approach with latent variable techniques for discrete data.
• Extended Mendoza-Blanco et al.'s simulation-based methods.
• Generalized Albert and Chib's methods for Bayesian analysis of binary response data.

Main Results:

• Developed a computationally manageable methodology for prevalence estimation.
• Successfully modeled the effects of covariates in prevalence estimation.
• Demonstrated the approach with real-world data examples.

Conclusions:

• The proposed method offers a more robust framework for prevalence estimation.
• Integrating test variability and covariate effects improves accuracy.
• This approach enhances Bayesian analysis for diagnostic accuracy studies.